There’s a moment that happens when you encounter a piece of media today a video, an audio clip, a photograph and before you process what it’s saying, you ask yourself: is this real?
That fraction of a second of doubt is what I think about when I think about deepfakes and democracy. People no longer trust what they see, even when it’s real. That erosion of trust is the real threat, and we’re closer to the point of no return than most are willing to admit.
The Liar’s Dividend
Most conversations about deepfakes focus on deception: someone sees a fabricated video and believes it to be false. The deeper, more corrosive problem runs in the opposite direction.
Legal scholars Bobby Chesney and Danielle Citron gave it a name: the liar’s dividend. When deepfakes become credible enough, anyone caught in genuine, damaging footage gains a new defense just claim it’s fake. The technology doesn’t need to fool audiences. It only needs to make denial sound reasonable.
I’ve watched this dynamic accelerate faster than most institutions were prepared to handle.
One MIT Media Lab study found that when people were simply told a photo might be AI-generated, their trust in all images significantly dropped. Not just the suspect image — all images. The Edelman Trust Barometer for 2025 found global trust in information sources at record lows. These aren’t isolated findings. They’re measuring the same underlying shift.
Once people stop trusting what they see, they start trusting less of everything.
We Got Here Faster Than Anyone Predicted
In 2017, a researcher from the University of Washington produced a synthetic video of Barack Obama. Experts warned of an impending crisis. Most observers assumed we had time.
We didn’t.
By 2024, an AI-generated voice of Joe Biden was calling New Hampshire voters, telling them to skip the primary. In Chicago, a fabricated audio clip of a mayoral candidate appeared to show him casually discussing police killing teenagers. In Slovakia, an audio deepfake of a leading presidential candidate discussing how to rig the election went viral two days before the vote.
Creating any of this no longer required a research lab or a nation-state budget. All it took were a few prompts and seconds.
The democratization of technology is the critical point. As Brookings Senior Fellow Darrell West put it, this is now “prop-driven and template-driven accessible to anyone.” The barrier between intent and execution has collapsed.
Detection has not kept pace. The structure of the problem makes this likely to hold: creation methods will consistently outpace detection methods. Every advance in identifying synthetic media creates training data for the next generation of synthetic media to evade it. It’s an arms race where the offense has the structural advantage.
What Democracy Actually Needs to Function
Democracy doesn’t run on votes alone. It rests on a shared epistemic foundation: a common understanding that certain things happened, that evidence matters, and that institutions have the legitimacy to act on that evidence.
Deepfakes attack this foundation at every level simultaneously.
Elections become vulnerable when any video can be denied. A candidate caught on camera no longer faces the weight of that footage; they can simply call it fake.
Accountability weakens when denial is cheap. Journalism slows down when editors are afraid to publish something that might turn out to be fabricated. The institutions we rely on to establish truth are working with rules written for a different era.
Across the 87 countries that held elections since 2023, 33 experienced deepfake-related incidents. No single deepfake has been confirmed to have decided an outcome. But the damage builds quietly as a growing sense that the information around us can’t be trusted.
And that sense changes everything. What people believe. What gets published. Who can be held accountable.
Why Regulation Arrives Late and Why That’s Not the Core Problem
The regulatory response has been real. The EU AI Act, passed in 2024, is the world’s first serious attempt at risk-based AI regulation. Several US states passed disclosure laws for AI-generated political content. The FCC proposed rules on the use of AI in broadcast political advertising. These represent genuine effort.
Regulation will always lag. By the time policies are written, the technology has already moved on. The EU AI Act was being debated while the tools it aimed to govern were releasing new versions quarterly. State-level laws in the US are inconsistent, constitutionally constrained, and difficult to enforce against anonymous actors.
But the deeper issue is a category mismatch. Regulation targets content, while the real problem is human perception. You can require disclosure labels on AI-generated political ads. You cannot require audiences to notice, process, or update their trust accordingly. The mechanism of harm operates at a psychological level that legal frameworks aren’t designed to reach.
This is why the coming decade’s most important battles over democratic integrity won’t be won primarily in legislatures.
Trust Is the Most Valuable Currency in Society Now
Every major technological era reorganizes what becomes scarce.
The industrial era: energy.
The internet era: attention.
In the AI era, authenticity is becoming a scarce resource, and scarcity produces value.
Here’s the paradox deepfakes create: by making everything potentially fakeable, they make verified truth more valuable than ever. The ability to prove something is real not just assert it, but prove it with traceable, verifiable evidence becomes a meaningful competitive advantage for everyone: media organizations, governments, companies, and individual leaders.
We’re already seeing the first attempts to rebuild trust through technology.
The C2PA standard adopted by LinkedIn, Adobe, and others embeds cryptographic metadata into media at the point of creation, so you can verify where content came from and whether AI was involved.
The EU’s Digital Services Act requires major platforms to label and trace synthetic content.
PwC and UNSW Sydney have been developing frameworks that treat authenticity as something you build into your systems, not something you assert after the fact.
The organizations that earn lasting trust in this environment will be the ones that can consistently prove what’s real. Credibility will belong to those who make truth traceable.
That’s both a challenge and an opportunity. The challenge anyone can now cheaply manufacture the appearance of credibility. The opportunity genuinely verifiable credibility becomes rarer, and therefore more powerful, than it has ever been.
What This Means Right Now
For leaders and organizations: your credibility is now a target. Deepfakes will be used to fabricate statements, manufacture crises, and create false evidence of misconduct. The question is whether you’ll have built the infrastructure to prove your reality when it happens: verified communication channels, content authentication protocols, and the institutional relationships that make your denials credible before you need to issue one.
For citizens: the goal of deepfake-driven disinformation isn’t always to make you believe something false. Often it’s to make you believe nothing to produce the kind of skepticism where disengagement feels rational. Checking out of the information environment entirely is the outcome that certain actors are specifically trying to produce.
Where This Leaves Us?
What gives me hope and what concerns me is the same thing. We are entering an era where trust is the most valuable currency in society. That means the stakes of building it, or destroying it, have never been higher. The organizations, institutions, and leaders who understand this early will have an advantage that compounds over time.
The point of no return isn’t a moment when deepfakes become perfect. It’s a moment when the erosion of trust becomes self-sustaining when doubt propagates without any further intervention. We may be closer to that point than most are willing to admit.
The time to build trust infrastructure is now, not after the next incident.
